The safety of AI mental health tools depends not only on the model but also on hidden systems that manage prompts, guardrails, retrieval, and human oversight.

Generative AI can identify ways around legal limits and safety filters governing mental health chats, raising new risks for users, lawmakers and AI companies.

The safety of AI mental health tools depends not only on the model but also on hidden systems that manage prompts, guardrails, retrieval, and human oversight.